Does my scan get uploaded anywhere?

No. Recognition runs in the tab: Tesseract.js with its WebAssembly core, plus pdf.js to draw PDF pages. The only network traffic is the one-off download of the OCR engine and your chosen language model; the document itself is read from local memory and never sent.

How many languages can it recognise?

The dropdown carries 38, from English, Spanish and German through Arabic, Hebrew, Hindi, Tamil, Thai, Japanese, Korean and both Chinese scripts. Each is shown with its Tesseract code so you can tell chi_sim from chi_tra.

Why is the extracted text full of mistakes?

OCR guesses characters from pixels, so anything that blurs the shapes hurts it: low resolution, JPEG artefacts, skew, shadows down the gutter of a book, coloured or textured paper, tight kerning, decorative fonts. A 300 dpi flat scan of clean print reads far better than a handheld photo of the same page.

Can it read handwriting?

It will try. The Handwritten mode tunes the engine to treat the page as one uniform block, which suits a page of notes, and the tool itself advises clear dark ink on light paper. Expect noticeably lower accuracy than printed text, and expect to proofread every line.

Does it work offline?

After the first run, largely yes. The page states the engine is roughly a 40 MB download that caches permanently, with each language adding about 1 to 5 MB. For languages served from GoChronos itself there is a "Save for offline" pill that fetches the model ahead of time.

Extract Text From a Scanned Document

Turn a scan or a photo of a page into editable text with browser-based OCR, then tidy the typos and pull out the key points. Nothing is uploaded.